AMD once again violating power specifications? (AMD RX-480)

[Coaxialgamer]

7 minutes ago, Coaxialgamer said:

 

 

[stconquest]

2 minutes ago, Briggsy said:

at 14nm it appears to.

How?

[LAwLz]

38 minutes ago, Trixanity said:

The other graphs in the source you refer to do list 50W average but if I recall with much bigger fluctuations than the RX 480. Since the 960 has a much lower floor, it'll drag the average number down regardless of what power it actually uses. That's how averages work. If the graph is to be trusted, it'll actually pull more than 75W and for more than just a tiny spike like the 225W spikes which are measured in what? Microseconds? So a second or two of pulling 75-90W for the 960 could be bad. Obviously it's worse to draw 82-84W (stock RX 480) over a minute. However that does not detract from the 960's problem. It's there. And its spikes are worse.

Don't refer to it as the "960's problem" because it was specifically the 960 Strix's problem.

The thing is that while spikes are bad, they are not as bad as continuous power draw. Like the PCPer article says, motherboards can handle spikes. Continuous draw over the specification is however a much bigger problem.

You are comparing apples and oranges. Abuse from spikes and constant abuse are not the same thing.

 

41 minutes ago, Trixanity said:

And for your second point. It seems like no one actually read the entirety of my post. I did clarify that it was only the Asus Strix card behaved that way. I put out the bait and you bit down on it.

Interesting. I made a headline pointing fingers at Nvidia and you rolled with it but the actual content clarified that Nvidia is not to blame, but Asus is. That means I'm a qualified tech journalist and I will go collect my diploma.

I did read your post. The problem is that, like we can see in this post, ignorant people will see your post and believe it. I saw that you were saying it was an Asus issue, but I have to spell it out to people and be extremely clear because otherwise someone with the same IQ as the calorie count of a Coke Zero will see your post and believe the bullshit.

 

I am not a fan of posting misinformation and then going "lol just kidding guys".

 

46 minutes ago, Trixanity said:

The Reddit post has been somewhat debunked by now. Someone misunderstood the spec.

Nvidia cards have a hard limit on what it can draw over the bus on recent cards. Where do you think that comes from? All I've read points to the vBIOS and not necessarily a hardware limit, although it could potentially also be "bottlenecked" by a weak power delivery. That means the problem can be fixed through a vBIOS update. Although it was still a mistake to split the load evenly between the bus and aux with the power consumption the 480 has. I will note that the components on the RX 480 are apparently beefier than anything Nvidia or AMD has previously made (reference) including Fury X, so I don't think AMD should fear running into power problems by cutting down on the power drawn through the bus (even though 3 phases are apparently linked to the bus and 3 to the aux). And finally, if it proves impossible to fix it in software, a recall is the only option. It won't burn out your system within a day but the clock is probably ticking.

Nobody but AMD knows if it can be fixed via a BIOS update. Ryan Shrout just said that it might be possible. Only time will tell. It might be a hardware issue that is causing it.

 

 

17 minutes ago, Blake said:

In breaking news, no-one in this thread has actually read the PCI spec to know this is not an issue. 

Have you read it? I have spent an unhealthy amount of time today reading though the most important parts. Can you please explain to me how this is a non-issue?

[zMeul]

5 minutes ago, stconquest said:

Except, GCN doesn't suck. 

it started showing it's age:

 

 

and GloFo's 14nm process didn't helped them either

[Briggsy]

3 minutes ago, stconquest said:

How?

480 pulls more power at 14nm than Maxwell did at 28nm (ie compared to 980), for less performance per watt. 

 

that sucks.

[Coaxialgamer]

7 minutes ago, Briggsy said:

They won't ditch GCN, AMD love to work a dead horse to hell and back. Just look at Bulldozer architecture, they were designing and running articles in magazines about bulldozer in 2006 and onward, and knew it sucked before they launched it.

The thing about  architectures is that they take a very  long time to design.  I am pretty sure amd would  have ditched bulldozer when they realized it wasn't competitive instead of releasing piledriver ( improved bulldozer),  steamroller and excavator if they had the choice.  The problem is that developing an architecture (whether for cpu or gpu)  takes  a lot of work.  Just look at nvidia and intel for example.  Broadwell is just a shrinked and slightly improved haswell.  Pascal is a shrinked maxwell. 

[Coaxialgamer]

4 minutes ago, zMeul said:

it started showing it's age:

 

 

and GloFo's 14nm process didn't helped them either

It technically is samsung's process,  so i don't think the manufacturing process is to blame here. 

[stconquest]

8 minutes ago, Briggsy said:

480 pulls more power at 14nm than Maxwell did at 28nm, for less performance per watt. 

 

that sucks.

@zMeul  (you too)

At these wattages, I could really not care less.  I want FPS.

 

They only shrunk the die and add some optimization and features like memory compression.  That lowers overall cost for them, and hopefully for us.  (hello Nvidia)

 

AMD still has AC engines that use more power than the Pascal architecture.  Nvidia is betting people won't care about this parallel computing feature because their design is so damn fast (high clocks).

 

GCN has better IPC still, and you will see that as these power issues get resolved. 

 

Although I can't recommend the ref. 480, I in no way would recommend a GTX 970... even at the same price.

[Briggsy]

3 minutes ago, stconquest said:

At these wattages, I could really not care less.  I want FPS.

 

They only shrunk the die and add some optimization and features like memory compression.

 

AMD still has AC engines that use more power than the Pascal architecture.  Nvidia is betting people won't care about this parallel computing feature because their design is so damn fast (high clocks).

 

GCN has better IPC still, and you will see that as these power issues get resolved. 

 

Although I can't recommend the ref. 480, I in no way would recommend a GTX 970... even at the same price.

That I can agree with. 970/980 is basically an EOL product now.

[zMeul]

6 minutes ago, LAwLz said:

Nobody but AMD knows if it can be fixed via a BIOS update. Ryan Shrout just said that it might be possible. Only time will tell. It might be a hardware issue that is causing it.

you can't re-route traces on a PCB with a BIOS update

RX480 has 6 power phases dedicated to the GPU and 1 aux

 

if you "route" power by disabling one or more of the phases, that immediately puts more load on the rest of the phases and it will suddenly increase the stress instead of normal load

 

 

[zMeul]

14 minutes ago, stconquest said:

@zMeul  (you too)

At these wattages, I could really not care less.  I want FPS.

 

They only shrunk the die and add some optimization and features like memory compression.  That lowers overall cost for them, and hopefully for us.  (hello Nvidia)

 

AMD still has AC engines that use more power than the Pascal architecture.  Nvidia is betting people won't care about this parallel computing feature because their design is so damn fast (high clocks).

 

GCN has better IPC still, and you will see that as these power issues get resolved. 

 

Although I can't recommend the ref. 480, I in no way would recommend a GTX 970... even at the same price.

that's not the issue

 

the issue is that a lot of people sold their 390s, 970s and even 980s expecting RX480 to be a godsend - guess what 

here, and I guess it's a general EU problem, RX480 price is very close to a custom cooled and OCed 970

[stconquest]

12 minutes ago, zMeul said:

you can't re-route traces on a PCB with a BIOS update

RX480 has 6 power phases dedicated to the GPU and 1 aux

 

if you "route" power by disabling one or more of the phases, that immediately puts more load on the rest of the phases and it will suddenly increase the stress instead of normal load

 

 

Depends how well they have built the controller (or just what controller they bought).  If a bios can limit total power draw through software, they can effectively lock out excessive voltages.  This mainly saves their(AMD) asses, but hurts anyone that bought the ref. design.  They (consumers), of course, will have the worst performers in the RX 480 family.

[stconquest]

2 minutes ago, zMeul said:

that's not the issue

 

the issue is that a lot of people sold their 390s, 970s and even 980s expecting RX480 to be a godsend - guess what 

here, and I guess it's a general EU problem, a RX480 is very close to a custom cooled and OCed 970

Although many people try to help out, some levels of stupidity have to be fixed client side. 

[laminutederire]

The RX 480 won't be a disaster once we get non reference cards I guess. The reference one isn't that bad, it should be good for a certain type of people though.

I heard on a  YouTube video, don't remember which one, that AMD may present a software fix, it is their product they know it better than we do, so we kinda should let them handle it.

Non reference cards should be alright with that, i fear that issue might delay them a bit though ! 

 

As for the debate around architecture, it takes time and MONEY, which they lack of, that's why they have struggled to make new architectures. But these new architectures may be coming after quite a while. At least it is with Zen on the CPU front, so who knows with GPU'S ? 

[Briggsy]

6 minutes ago, laminutederire said:

The RX 480 won't be a disaster once we get non reference cards I guess. The reference one isn't that bad, it should be good for a certain type of people though.

I heard on a  YouTube video, don't remember which one, that AMD may present a software fix, it is their product they know it better than we do, so we kinda should let them handle it.

Non reference cards should be alright with that, i fear that issue might delay them a bit though ! 

 

As for the debate around architecture, it takes time and MONEY, which they lack of, that's why they have struggled to make new architectures. But these new architectures may be coming after quite a while. At least it is with Zen on the CPU front, so who knows with GPU'S ? 

I believe Raja had no part in the making of polaris and Vega. His baby is Navi, which will probably see some sweeping changes in power draw and design. Unfortunately, people have their hearts set on Vega, and we have no idea what's in store there. Raja certainly would dump it if he could and push Navi forward. after all, he's pushing someone else's design right now and trying to be enthusiastic about something he didn't design himself.

[3DOSH]

Some people fail to see that AIB cards are coming the next few weeks,  those well be clocked higher, given robust power delivery and cooling. They well be in 300$ range .  We saw the RX doing 1400 mhz easy on tower cooler,  now add to that 8 pin and 6pin pcie for power we well have +25% more performance easy.  

Also AMD said 500$ performance in dx12 games not 11 and you can go back and check, people just hyped the card to much . People buying 970s are frankly stupid, just look at dx12 benchmarks and see how it demolishes the 970 and goes head to head with 980 and that's at stock.

Be smart and decided if you rather have today's performance for today's games or tomorrow's performance for future games. 

[Blake]

37 minutes ago, LAwLz said:

 

Have you read it? I have spent an unhealthy amount of time today reading though the most important parts. Can you please explain to me how this is a non-issue?

go read https://www.reddit.com/r/Amd/comments/4qmlep/rx_480_powergate_problem_has_a_solution/

 

And clearly you haven't read it. otherwise you would also know this is a nothing story and never was, but the usual "DRAMA ALERT" that gets thrown around every time there is a new product release or some other polarised topic mentioned.

[79wjd]

@3DOSH it will be a while before dx12 performance really starts to outpace dx11 performance in newer titles. The bigger reason why I would go for the 480 over 970 is because the 970 won't have the longevity of the 480 as it's very driver dependent. 

[laminutederire]

5 minutes ago, Briggsy said:

 

Navi is a bit far away though. He needs to secure stable incomes with polaris and Vega in order to have money to develop Navi, that's why he'll do what he can to create value from the designs. I think polaris isn't such a disaster. The power draw issue aside, the rx 480 is still better than what there was in the 300 series (except the 390X, but non reference designs might close that gap with better cooling and more power than the reference design but less than the 390X)

Will Navi even have silicon transistors by the way?

[Briggsy]

6 minutes ago, 3DOSH said:

Some people fail to see that AIB cards are coming the nest few weeks,  those well be clocked higher, given robust power delivery and cooling. They well be in 300$ range.  We saw the RX doing 1400 mhz easy on tower cooler,  now add to that 8 pin and 6pin pcie for power we well have +25% more performance easy.  

Also AMD said 500$ performance in dx12 games not 11 and you can go back and check. People buying 970s are frankly stupid, just look at dx12 benchmarks and see how it demolishes the 970 and goes head to head with 980 and that's at stock.

Be smart and decided if you rather have today's performance for today's games or tomorrow's performance for tomorrow's games. 

few things.

 

1 - The tower cooler test reached 1400mhz stable on one card, and on three other cards they could only do 1330-1350 with the test, so its still a lottery.

2 - the power draw to hit 1400mhz with an 8pin connector will likely be 200-250 watts depending on the card.

3 - My memory is good enough i don't have to check, AMD said the 480 would deliver VR performance similar to a 500 dollar graphic card.

4 - the 480 barely keeps up with reference 970 (although i have never and will never recommend a 970)

[3DOSH]

2 minutes ago, djdwosk97 said:

@3DOSH it will be a while before dx12 performance really starts to outpace dx11 performance in newer titles. The bigger reason why I would go for the 480 over 970 is because the 970 won't have the longevity of the 480 as it's very driver dependent. 

Not really that long,  Battlefield 1, deus ex and more are coming out at the end of the year so not that much to wait also with dx12 driver support is less dependent on AMD,  this goes hand in hand with driver maturity. 

[79wjd]

1 minute ago, 3DOSH said:

Not really that long,  Battlefield 1, deus ex and more are coming out at the end of the year so not that much to wait also with dx12 driver support is less dependent on AMD,  this goes hand in hand with driver maturity. 

I'm not saying dx12 titles aren't coming, but unless I'm mistaken those all support dx11, and the dx11 performance on nvidia cards will probably be fairly similar to the dx12 performance on AMD cards as has been the case in the past.

[laminutederire]

1 minute ago, djdwosk97 said:

I'm not saying dx12 titles aren't coming, but unless I'm mistaken those all support dx11, and the dx11 performance on nvidia cards will probably be fairly similar to the dx12 performance on AMD cards as has been the case in the past.

Well that's good news that means Nvidia doesn't crush AMD

 

[3DOSH]

8 minutes ago, Briggsy said:

few things.

 

1 - The tower cooler test reached 1400mhz stable on one card, and on three other cards they could only do 1330-1350 with the test, so its still a lottery.

2 - the power draw to hit 1400mhz with an 8pin connector will likely be 200-250 watts depending on the card.

3 - My memory is good enough i don't have to check, AMD said the 480 would deliver VR performance similar to a 500 dollar graphic card.

4 - the 480 barely keeps up with reference 970 (although i have never and will never recommend a 970)

1,2 are up for debate but we'll find out soon enough. 3 I was generally talking,  I don't know why you took it personally. 4 not all people have the foresight to pick a card,  they just look at the benchmarks and decide when the benchmarks don't tell the whole story. 

[patrickjp93]

3 hours ago, stconquest said:

If you never test, all the back and forth means shit. 

Nope. Mathematics is proof of that.